Re: [PATCH] mm: use managed_zone() for more exact check in zone iteration

[Andrew Morton]

On Thu, 15 Nov 2018 07:50:40 +0800 Wei Yang <richard.weiyang@xxxxxxxxx> wrote:

> For one zone, there are three digits to describe its space range:
> 
>     spanned_pages
>     present_pages
>     managed_pages
> 
> The detailed meaning is written in include/linux/mmzone.h. This patch
> concerns about the last two.
> 
>     present_pages is physical pages existing within the zone
>     managed_pages is present pages managed by the buddy system
> 
> >From the definition, managed_pages is a more strict condition than
> present_pages.
> 
> There are two functions using zone's present_pages as a boundary:
> 
>     populated_zone()
>     for_each_populated_zone()
> 
> By going through the kernel tree, most of their users are willing to
> access pages managed by the buddy system, which means it is more exact
> to check zone's managed_pages for a validation.
> 
> This patch replaces those checks on present_pages to managed_pages by:
> 
>     * change for_each_populated_zone() to for_each_managed_zone()
>     * convert for_each_populated_zone() to for_each_zone() and check
>       populated_zone() where is necessary
>     * change populated_zone() to managed_zone() at proper places
> 
> Signed-off-by: Wei Yang <richard.weiyang@xxxxxxxxx>
> 
> ---
> 
> Michal, after last mail, I did one more thing to replace
> populated_zone() with managed_zone() at proper places.
> 
> One thing I am not sure is those places in mm/compaction.c. I have
> chaged them. If not, please let me know.
> 
> BTW, I did a boot up test with the patched kernel and looks smooth.

Seems sensible, but a bit scary.  A basic boot test is unlikely to
expose subtle gremlins.

Worse, the situations in which managed_zone() != populated_zone() are
rare(?), so it will take a long time for problems to be discovered, I
expect.

I'll toss it in there for now, let's see who breaks :(